Transportation barrier data trust: what claims-based proxy measures miss
Claims-based proxy measures for transportation barriers misclassify up to 40% of patients who actually face access problems. Area-level deprivation indices, ZIP code distance calculations, and missed appointment flags each introduce distinct distortions that compound when fed into population health AI. Building a transportation barrier data trust requires patient-level verification, temporal specificity, and provenance scoring that proxy data cannot provide.
Claims data identifies a patient who missed three appointments. It does not tell you whether the patient lacked a car, whether the bus route was cancelled that week, whether the appointment was at 7:30 AM and the first available ride-share opens at 9, or whether the patient simply chose not to go. Yet most health systems treat missed appointment flags as transportation barrier signals. The gap between that proxy and reality is where population health AI breaks down.
Transportation barriers account for 25% or more of missed clinic appointments, according to CMS. An estimated 3.6 million Americans delay or forgo medical care each year because they cannot get to a provider. These numbers are themselves approximations, built from survey data and claims-derived inferences. The actual prevalence is almost certainly higher, because the measurement instruments systematically undercount the populations most affected.
What claims-based proxy measures actually capture
The standard approach to identifying transportation barriers from claims data relies on a small set of signals: missed appointment rates, geographic distance between patient address and provider location, Z59 ICD-10 codes for problems related to housing and economic circumstances, and area-level indices like the Area Deprivation Index (ADI) or Social Vulnerability Index (SVI).
Each of these proxies captures something real. None of them captures the thing they claim to measure.
Missed appointment rates conflate transportation barriers with scheduling conflicts, childcare gaps, health literacy issues, fear of medical bills, and simple forgetfulness. A 2022 study in Health Affairs found that only 28% of patients flagged as "no-show" cited transportation as the primary reason. The remaining 72% had other barriers or no identifiable barrier at all.
Z59 codes are inconsistently applied. Coding for social determinants depends on whether the clinician asks, whether the patient discloses, and whether the coder selects the Z code alongside the clinical diagnosis. In practice, Z59 code capture rates vary from 1.2% to 14% across health systems for comparable patient populations. That variation reflects documentation practice, not actual prevalence.
Geographic distance calculations assume a straight line or driving route between a patient's address on file and the provider location. They do not account for public transit availability, transit schedule reliability, road conditions, or whether the patient has access to a vehicle. A 5-mile trip in Manhattan is a fundamentally different barrier than a 5-mile trip in rural Mississippi.
Why area-level indices fail at patient-level prediction
The ADI and SVI assign deprivation scores to census tracts or block groups. These scores correlate with transportation access at the population level. They fail at the individual level.
A census tract with a high ADI score contains both residents who own cars and residents who do not. A tract with a low ADI score may include elderly residents who no longer drive, patients with mobility limitations, or households where one car serves a family of five with competing schedules. The ecological fallacy, applying group-level statistics to individuals, is the core problem with area-level SDOH proxies.
As we have written about extensively, geocoding accuracy and the census tract mismatch problem introduce additional error. Patient addresses in claims data are frequently outdated. Between 11% and 15% of Medicaid beneficiaries change addresses annually without updating their enrollment records. When a patient's address maps to the wrong census tract, the ADI or SVI score assigned to them describes a neighborhood where they do not live.
This compounding error, wrong address plus wrong tract plus ecological fallacy, means that area-level transportation barrier proxies can misclassify 30% to 40% of individual patients. That is not a rounding error. That is a data quality failure that propagates through every downstream model.
What SDOH screening instruments add and what they miss
The AHC-HRSN (Accountable Health Communities Health-Related Social Needs) screening tool includes a direct question about transportation: "In the past 12 months, has lack of reliable transportation kept you from medical appointments, meetings, work, or from getting things needed for daily living?"
This is closer to ground truth than any claims-derived proxy. It asks the patient directly. But screening data has its own trust problems.
Screening completion rates for transportation questions range from 45% to 78% across health systems that have adopted the AHC-HRSN. Non-completion is not random. Patients who face transportation barriers are less likely to be present for screenings, which are typically administered during clinical encounters. The patients most affected are the ones most likely to be absent from the data.
Screening responses are point-in-time. A patient who had reliable transportation in January may lose vehicle access in March due to a breakdown, job loss, or household change. The 12-month lookback window in the AHC-HRSN question captures historical experience but does not reflect current status. For AI models predicting near-term appointment adherence or care gap closure, a screening response from six months ago is already stale.
The AHC-HRSN instrument quality requirements we have previously documented apply with particular force to transportation screening. Instrument administration context, response completeness, and temporal validity all affect whether a screening response can be trusted for clinical or operational use.
Transportation barrier examples that proxy data cannot see
The gap between proxy measurement and lived reality shows up in specific, concrete ways that claims data never captures.
A patient in rural Appalachia lives 22 miles from the nearest specialist. Claims data shows the distance. It does not show that the only road floods three times per year, that the patient's truck needs $800 in repairs, or that the patient's spouse works the day shift and cannot provide a ride before 5 PM.
A Medicaid beneficiary in Houston has a bus route that technically connects her neighborhood to the county hospital. The trip requires two transfers, takes 90 minutes each way, and the return bus stops running at 6 PM. Her dialysis appointments end at 4:30 PM three days per week. On paper, she has transit access. In practice, she misses one session every two weeks.
An elderly patient in suburban Chicago stopped driving after a minor accident. Her address has not changed. Her ADI score has not changed. Her claims show increasing appointment gaps. No proxy measure identifies the actual cause: she is afraid to drive and embarrassed to ask for help.
These are not edge cases. They represent the majority of transportation barriers. The community health organizations and SDOH data quality trust gap exists precisely because the organizations closest to these patients, community-based organizations, collect the richest transportation barrier data but have the weakest integration with health system data infrastructure.
Solutions to transportation barriers in healthcare require better data, not just better rides
Most health system responses to transportation barriers focus on the intervention side: ride-share partnerships with Lyft or Uber Health, non-emergency medical transportation (NEMT) benefit coordination, and mobile health units. These interventions matter. But they are being deployed against a faulty map.
When a health system uses proxy data to identify which patients need transportation assistance, it systematically over-serves some populations and under-serves others. Patients flagged by area-level indices may receive ride-share offers they do not need. Patients whose barriers are invisible to claims data receive nothing.
The COVID pandemic made this worse. During 2020 and 2021, transportation barriers shifted dramatically as public transit reduced service, ride-share availability dropped, and patients avoided shared vehicles. Claims-based proxy measures, which rely on historical patterns, could not adapt to these rapid changes. Health systems that depended on pre-pandemic ADI scores to target transportation interventions were operating on data that described a world that no longer existed.
A transportation barrier data trust requires data that is verified at the patient level, timestamped with recency metadata, and scored for reliability before it enters any model or workflow.
Key statistics
What a transportation barrier data trust actually requires
Building trustworthy transportation barrier data means scoring every data element across dimensions that proxy measures ignore.
Provenance. Where did this transportation barrier flag originate? A patient self-report during an AHC-HRSN screening has different provenance than an inferred flag from a missed appointment pattern. A community health worker's home visit assessment has different provenance than a census-tract-level ADI assignment. Each source carries different reliability, and any system that treats them equivalently is producing untrustworthy outputs.
Recency. Transportation access changes faster than almost any other social determinant. A car breaks down. A bus route is cancelled. A household member who provided rides moves out. Seasonal weather makes rural roads impassable. Transportation barrier data older than 90 days should be treated as provisional. Data older than 180 days should trigger re-verification before clinical use.
Concordance. When multiple data sources describe the same patient's transportation status, do they agree? If a screening says "no barrier" but claims show a pattern of missed appointments concentrated on days when no public transit runs, the concordance failure is itself a signal.
Validation. Has the transportation barrier data been confirmed through a second source? A patient-reported barrier validated by a community health worker assessment is fundamentally more trustworthy than either source alone.
The DTI framework scores these dimensions explicitly. Provenance accounts for 25% of the total trust score. Recency accounts for 15%. Concordance and Validation each account for 10%. For transportation barrier data, where temporal volatility is high and proxy contamination is endemic, these weightings reflect real-world risk.
How proxy data limitations propagate through health AI
When population health AI models train on transportation barrier proxy data, they inherit every distortion in that data.
A readmission prediction model that uses ADI-derived transportation flags will learn that patients in high-deprivation census tracts are readmission risks. This correlation is real at the population level. But the model cannot distinguish between patients in those tracts who have reliable transportation and those who do not. It will generate false positives for patients who happen to live in high-ADI areas but face no transportation barrier, and false negatives for patients in low-ADI areas who cannot get to follow-up appointments.
As we have documented in our work on readmission prediction model bias, training data trust directly determines clinical AI performance. Transportation barrier proxy data with a DTI score below 40 should not be used for model training without explicit acknowledgment of its limitations and appropriate uncertainty quantification.
The claims data lag problem compounds this further. Even if a claims-based transportation proxy were accurate at the time of the encounter, the 30-90 day delay before claims data is available means the signal arrives too late to inform real-time interventions.
The CBO data bridge
Community-based organizations collect the highest-fidelity transportation barrier data in healthcare. A community health worker who visits a patient's home and observes that there is no car in the driveway, that the nearest bus stop is a mile away on a road without sidewalks, and that the patient mentioned their neighbor sometimes drives them to the pharmacy, has captured richer and more accurate transportation data than any claims-derived proxy.
The problem is that CBO data rarely reaches health system data infrastructure in structured, standardized form. When it does, it arrives without provenance metadata, without timestamp precision, and without the quality scoring that would allow it to be weighted appropriately against other data sources.
The CBO data trust requirements for Medicaid value-based programs we have outlined previously describe exactly this challenge. CBO-collected transportation data needs the same trust scoring infrastructure as clinical data. Without it, the richest source of ground truth about transportation barriers remains locked outside the data systems that make intervention decisions.
What the CMS ACCESS program means for transportation data
The CMS ACCESS program, which provides up to $420 per beneficiary per year for community health integration services, creates a financial incentive to collect and act on SDOH data including transportation barriers. But the program's data requirements also raise the stakes for data trust.
Health systems claiming ACCESS payments must demonstrate that they screened for social needs and connected patients to community resources. If the screening data is incomplete, if the transportation barrier flags are proxy-derived rather than patient-verified, or if the intervention documentation lacks provenance, the claims become audit targets.
The CMS ACCESS data foundation requirements we have described make clear that compliance pressure is moving toward verifiable, scored data. Transportation barrier data that cannot demonstrate its provenance and recency will not survive regulatory scrutiny.
Building the trust layer for transportation data
The path forward is not to abandon proxy measures entirely. Area-level indices, distance calculations, and missed appointment patterns all contain signal. The path forward is to score each data source for trustworthiness, weight it appropriately, and never treat a proxy as ground truth.
A transportation barrier data trust scores every input: the ADI value and its geocoding accuracy, the screening response and its administration context, the missed appointment pattern and its temporal specificity, the CBO assessment and its recency. Each element receives a DTI score. The composite picture, built from scored components, gives population health teams and AI models a calibrated view of transportation access that proxy data alone cannot provide.
This is not a theoretical framework. It is the operational requirement for any health system that wants to deploy AI against SDOH data and have the outputs mean something.
The DTI Engine scores every health data record 0-100 across 8 trust dimensions before your AI model sees it. For health systems building transportation barrier interventions, SDOH-informed care models, or population health AI, trust-scored data is the difference between targeting the right patients and targeting the convenient ones. If your team is evaluating SDOH data for training, compliance, or clinical use, schedule a conversation with the SuperTruth commercial team or (215) 918-4140.
Further reading:
Jason Alan Snyder
Co-founder of SuperTruth and Artists & Robots, and an inventor on the Data Trust Index patents. Twenty-plus years building technology inside Interpublic Group. He writes here nearly every day on data trust, provenance, and what AI should be allowed to act on, and publishes essays on his Substack.
About SuperTruth · LinkedIn · Substack · jasonalansnyder.com
See it in practice
DTI scores the record, not the patient.
8 dimensions. 0 to 100. Travels with every record permanently.